Claude learned to create videos with dynamic UI

The new Claude model has demonstrated advanced motion design skills, learning to generate stylish videos with dynamic interfaces. The artificial intelligence is capable of assembling animations, selecting a background track, and editing the final video on its own.

To create a video, one HTML file with a size of 1440x1440 pixels is used, where all styles are calculated based on time within the seek(t) function. In the process, the model uses numpy to analyze the music track under the beat grid, and video rendering is performed through Playwright with subsequent processing in ffmpeg to add a motion blur effect at 60 frames per second.

The animation script is built on a seamless transformation of one figure without stitching: the element changes shape, radius, and color, and the content switches with a slight blur. The cursor controls the changes using real clicks and drags, and the final frame completely matches the initial one to create an infinite loop.

This approach allows for the automation of creating high-quality interface demonstrations without using template solutions and unnecessary visual effects.